ai-cur Sign up

Know which AI tools earn their seat.

The seats nobody opened, the spend beside each merged pull request, and a budget that stops the overrun this month. Claude Code, Codex, Cursor, Copilot and the API keys behind your agents, on one page.

Team view for the Platform team in the ai-cur demo: each member with settled cost, merged pull requests, changed lines and cost per merged pull request for August 2026
A team lead's own team: what each person cost and what they merged, in the month it happened.
Reads AnthropicOpenAICursorGitHub CopilotGitHubJiraOktaYour own exports
What you get

Four answers a lead cannot get from the vendors.

Each vendor shows you its own bill. None of them shows your people, your teams, or the work that merged while the meter ran. ai-cur reads all of them and joins the spend to that.

  1. Who is using what

    Every seat and every API key resolved to a person and a team, across Claude, OpenAI, Cursor and Copilot, in one month view.

  2. What it cost, settled

    Charges as they were actually bought: seat, usage, credit or commitment, held open until the vendor closes them, never added across currencies.

  3. What merged beside it

    Pull requests from GitHub and issues from Jira, joined to the same people in the same month, so spend sits next to output.

  4. What to keep, cut or cap

    Idle seats to drop, plans that no longer fit, and a budget with a cap your gateway enforces before the month gets away.

What it reads and what you declare

Seats

Find the seats nobody opened.

Every paid seat is checked against its own vendor's usage for the month and marked active, idle or not measurable. With no vendor usage to measure against, a seat reads not measurable, never idle.

In the demo tenant, two people left in July and kept their seats: August shows 280.00 USD of idle seats.

Idle seats and plan fit

Paid seats and measured activity in the ai-cur demo: seats per team and provider with idle, active and not measurable counts, seat cost and idle cost for August 2026
Seats per team and provider, with the idle ones and what they cost, nobody named.
Work

Put the spend next to what merged.

Cost per merged pull request for a person, a team or the company, printed with the cost and the count it came from. It puts the spend beside the output; whether the tool earned it is your call, made with the number in front of you.

A team smaller than your comparison minimum is not ranked against teams its manager does not run.

Cost per pull request, issue and project

Measured work for August 2026 in the ai-cur demo: cost per merged pull request with its numerator, denominator and coverage, and cost per resolved issue shown as unavailable
Cost per merged pull request with the count it came from; a source that is off reads unavailable, never zero.
Budgets

Catch the overrun while the month is open.

Give a team, a project or a person a monthly budget with alert thresholds, then add soft and hard caps. ai-cur publishes the decision for your gateway to read, and your gateway does the blocking.

In the demo, Platform has spent 2,676.35 USD against its 2,400.00 USD August budget, and the panel says so.

Budgets, caps and the cap decision

Budgets panel in the ai-cur demo: Platform, August 2026, a 2,400.00 USD budget with 2,676.35 USD consumed, 112% of budget, alert thresholds at 80 and 100 percent
A team over budget with the month still open, and the thresholds that said so first.
Start here

Start free, on your own Mac, today.

The AI Usage Tracker is open source and free for good. It sits in your menu bar, reads your Claude Code and Codex sessions where they already live on your machine, and can watch your Claude, Codex and Cursor plan limits.

No account and no card. When the team wants a shared view, each install is enrolled with its own revocable token, and its figures join the team's page.

The Tracker on github.com/harvto-llc/ai-usage-tracker

One engineer's cost per merged pull request by month, June to August 2026, in the ai-cur demo, with settled cost and merged pull request counts
One engineer's own months: cost beside merged pull requests, visible to nobody else.
Finance

Hand finance a file, not a debate.

The finance view closes a month by cost centre and provider, holds unsettled charges back, and exports a CSV whose total is the page's total.

In the demo, the August file comes to 9,361.08 USD, the figure on the page above it.

The finance month

Export for posting in the ai-cur demo finance view: a download for settled cost for August 2026 in USD, total in the file 9,361.08 USD, and the columns the file carries
The month closed by cost centre and provider, and the file whose total is the page's total.

And the rest of what it answers.

Each one has its own section, with the screen that answers it, on the product page.

Three ways to start.

Free

The whole Tracker for one engineer, and the hosted view of your own reports. Team totals appear as teammates install.

Team

For the lead who answers for the spend: idle seats, plan fit, budgets, caps, alerts and exports, with a trial and no card.

Business

For larger estates: your own identity provider, declared contracts and commitments, and the Databricks, ServiceNow or on-prem profile.

What each includes, and what it costs

Start on your own numbers. Bring the team when it is ready.

Sign up with a work email, Google or Microsoft. The Tracker and your own view cost nothing, and a team gets sixty days of Team before anyone is asked for a card.

One question to bring: what does your company spend on AI tools per employee each month? The answer says whether seats or budgets deserve the first look.